A multi-objective weighted voltage management and power configuration method

Abstract

The application discloses a multi-target weighted voltage management and power configuration method, and relates to the technical field of power quality management of distribution networks; the application firstly constructs a basic voltage-reactive power sensitivity matrix through a Newton-Raphson method, and obtains a voltage-reactive power sensitivity matrix that is in line with actual working conditions through double-dimensional correction of equipment aging time dimension and topological coupling space dimension, and simultaneously constructs a multi-dimensional resource portrait; then, feature purification is completed through an improved lightweight convolutional neural network, and data credibility verification is completed in combination with blockchain technology; subsequently, a leading disturbance source of the distribution network is identified, and multi-target optimization weights are dynamically allocated, so that adaptation and matching of scene demand and resource capacity are completed; finally, a power configuration scheme and a voltage management strategy are output through GA-PSO hybrid intelligent algorithm optimization solution; the application greatly improves the accuracy and working condition adaptability of voltage management, and takes into account management effect and operation economy.

Description

Technical Field

[0001] This invention relates to the field of power quality management technology for distribution networks, specifically a method for multi-objective weighted voltage management and power configuration. Background Technology

[0002] With the large-scale grid connection and application of new energy equipment such as distributed photovoltaics and user-side energy storage in distribution networks, the traditional passive unidirectional power supply distribution network is gradually being transformed into an active network with bidirectional interaction between source and load. The dynamics and uncertainties of the distribution network operation conditions have been significantly improved, and the difficulty of managing power quality problems such as voltage fluctuations and three-phase imbalances has continued to increase. Precise and efficient voltage management and power configuration have become the core requirements for ensuring the safe, stable, economical and efficient operation of the active distribution network.

[0003] Current technologies related to voltage management in distribution networks typically involve constructing voltage-reactive power sensitivity analysis models based on power flow calculations, and combining these with intelligent optimization algorithms to perform multi-objective power configuration optimization. This enables the coordinated scheduling of voltage regulation and reactive power resources in distribution networks, achieving certain management effects in traditional passive distribution network scenarios.

[0004] However, in the scenario of active distribution networks with high penetration of distributed renewable energy, the voltage-reactive power analysis models used by these technologies are mostly based on the ideal operating state of equipment and the construction of a fixed power grid topology. They do not fully match the dynamic changes in equipment performance and the real-time adjustment of the topology during the actual operation of the distribution network. This results in a significant deviation between the model analysis results and the actual operating state of the distribution network, which in turn leads to insufficient accuracy of voltage management strategies and power configuration schemes. They are difficult to adapt to the complex and ever-changing operating needs of active distribution networks and cannot guarantee the power quality and operating economy of the distribution network in the long term.

[0005] Therefore, it is necessary to design a multi-objective weighted voltage governance and power configuration method. Summary of the Invention

[0006] The purpose of this invention is to provide a multi-objective weighted voltage management and power configuration method to solve the problem that the existing voltage-reactive power analysis model mentioned in the background art deviates greatly from the actual operating conditions of the distribution network, resulting in insufficient accuracy of voltage management strategies and power configuration schemes and poor adaptability to operating conditions.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for multi-objective weighted voltage governance and power configuration, comprising:

[0008] Acquire distribution network topology data, node electrical data, real-time distribution network topology information, and governance resource operation data; construct a voltage-reactive power sensitivity matrix based on distribution network topology data and node electrical data.

[0009] Based on the governance resource operation data and the real-time topology information of the distribution network, the voltage-reactive power sensitivity matrix is ​​corrected in both time and space dimensions. The node sensitivity is determined by the corrected voltage-reactive power sensitivity matrix, and the power injection constraints are determined based on the load type in the node electrical data. At the same time, a multi-dimensional resource profile is constructed based on the governance resource operation data.

[0010] Based on the corrected voltage-reactive power sensitivity matrix, nodal electrical data, and multi-dimensional resource profile, an original feature dataset is generated.

[0011] The original feature dataset is purified and its credibility is verified to obtain a feature dataset containing scene features and resource features.

[0012] Identify the dominant disturbance sources in the distribution network based on scenario characteristics, dynamically allocate multi-objective optimization weights based on the dominant disturbance sources and node sensitivity, and adapt and match scenario requirements with resource capabilities based on scenario characteristics and resource characteristics to generate resource combination recommendation results.

[0013] Using multi-objective optimization weights as the priority of optimization objectives and resource combination recommendation results as the configuration objects, combined with power injection constraints, multi-objective power configuration optimization is performed to generate power configuration schemes and output corresponding voltage management strategies.

[0014] As a further technical solution of the present invention, the power distribution network topology data includes the electrical parameters of the power distribution network lines, node connection relationships, and branch switch status;

[0015] The node electrical data includes the total number of distribution network nodes, node type, rated voltage, load type, and rated parameters;

[0016] The real-time topology information of the distribution network includes the real-time topology structure of the distribution network, line switching and switch operation information;

[0017] The governance resource operation data includes basic ledgers, real-time operation data, historical performance data, and equipment status information of the governance equipment on the power grid side and the user side.

[0018] The voltage-reactive power sensitivity matrix is ​​constructed using the Newton-Raphson power flow calculation method: First, the input distribution network topology data and node electrical data are standardized and preprocessed to form a power flow calculation dataset; initial values ​​for node voltage and phase angle are set for iteration, and the iteration convergence criterion is determined based on the node voltage deviation and phase angle deviation between adjacent iterations; the iteration loop is entered: using the node voltage, phase angle, and the power flow calculation dataset of the current iteration as input, the power deviation between the actual power and the given power of each node is calculated; using the current node voltage and phase angle as input, the Jacobian matrix of the power flow equation is constructed, and the sub-blocks associated with reactive power-voltage are retained; for the... The Jacobi matrix is ​​inverted to obtain the correction amount for the node voltage and phase angle. The current node voltage and phase angle are updated with the correction amount, and the updated node voltage and phase angle are compared with the judgment criteria. If the node voltage deviation and phase angle deviation of two adjacent iterations do not meet the judgment criteria, the updated node voltage and phase angle are returned to the power deviation calculation step as the current value of the next iteration until the judgment criteria are met. Finally, the reactive power-voltage correlation sub-blocks retained in the converged Jacobi matrix are inversely operated to generate a voltage-reactive power sensitivity matrix. The matrix elements are used to characterize the voltage change amplitude of the corresponding node when a unit reactive power is injected into a single node.

[0019] As a further technical solution of the present invention, the time dimension correction is achieved by constructing an equipment health assessment model: based on the equipment status information in the governance resource operation data, three types of status indicators are extracted: equipment operating years, output current deviation rate, and real-time temperature rise, and an equipment health assessment model is constructed; corresponding weights are set according to the degree of influence of each indicator on equipment performance degradation, and the equipment health assessment model quantifies the degree of performance degradation of the equipment in long-term operation, and calculates the equipment health; then, the equipment health is used as an aging degradation factor to adapt and correct each element of the sensitivity matrix.

[0020] The spatial dimension correction is a topology coupling correction, which is achieved through the branch impedance coupling coefficient: first, the branch node group of the same line is identified through the real-time topology information of the distribution network, and the branch impedance coupling coefficient is calculated; then, the sensitivity matrix after the time dimension correction is corrected a second time.

[0021] After completing the time and space dual-dimensional correction, based on the corrected voltage-reactive power sensitivity matrix, the absolute value of the voltage response intensity of each node to reactive power injection is calculated as the node sensitivity of that node. Nodes with node sensitivity exceeding the preset threshold are marked as high-sensitivity nodes and are used as the targets for priority allocation of subsequent governance resources. The preset threshold is set according to the voltage governance accuracy requirements of the distribution network.

[0022] The power injection constraints are set according to the characteristics of the load type. For resistive, inductive, and capacitive loads, the upper limit of active power injection, the range of reactive power adjustment, and the priority adjustment direction of the corresponding nodes are determined respectively. The power injection constraints and the modified sensitivity matrix together serve as the constraints for subsequent power configuration optimization.

[0023] As a further technical solution of the present invention, the multi-dimensional resource profile is constructed based on the basic ledger, real-time operation data, and historical performance data in the governance resource operation data, and adopts a four-dimensional structured architecture of basic attributes, real-time operation status, historical performance, and adaptation scenarios. The basic attributes are generated based on the basic ledger and include fixed static data such as resource type, resource ID, access node, device rated capacity, and adjustable power factor range. The real-time operation status is obtained based on the real-time operation data and includes dynamically updated data on response speed, current adjustment margin, and operation status. The historical performance is generated based on historical performance data and includes full lifecycle statistics such as device adjustment accuracy, fault frequency, and fault-free operation time. The adaptation scenarios are preset tag data, including the disturbance types that the device prioritizes for adaptation, the sensitivity level of the adaptation node, and the cost threshold.

[0024] The current adjustment margin in the real-time operating status serves as a resource adjustment margin constraint for power configuration optimization; the rated capacity and adjustable power factor range of the equipment in the basic attributes serve as equipment safe operation constraints for power configuration optimization.

[0025] The statistical data in the historical performance and the tag data in the adapted scenario constitute the resource profile tag;

[0026] Establish a supporting dynamic update mechanism to update real-time operating status data at a preset fixed short cycle and iterate historical performance data at a preset fixed long cycle; the short cycle is set to 10 minutes based on the frequency of power distribution network condition monitoring, and the long cycle is set to monthly based on the accumulation cycle of equipment operation statistics.

[0027] As a further technical solution of the present invention, the original feature dataset is divided into two categories: a scenario original feature subset and a resource original feature subset; wherein the scenario original feature subset is composed of the node sensitivity determined by the modified voltage-reactive power sensitivity matrix and the load type in the node electrical data, and the resource original feature subset is composed of the full index data of the multi-dimensional resource profile.

[0028] As a further technical solution of the present invention, the feature purification is implemented using an improved lightweight neural network: first, the original feature dataset is standardized and preprocessed; then, the improved lightweight neural network, which has been pre-trained, extracts core scene features and core resource features through parallel branches: the scene feature branch extracts core scene features related to the power distribution network conditions, disturbance types, and governance requirements, and removes redundant scene features that are not directly related to the voltage governance effect; the resource feature branch extracts core resource features related to the regulation capability, adaptability, and economy of governance equipment, and removes redundant resource features that are not directly related to resource scheduling; the improved lightweight neural network, which has been pre-trained, serves as the feature extraction model.

[0029] The credibility verification is achieved through blockchain distributed ledger technology: First, a traceability tag containing the collection device identifier, timestamp, digital signature, original data, and hash value is attached to each purified feature data; then, tamper verification is performed to compare the consistency between the current data and the blockchain evidence information. If they are inconsistent, the historical compliant data of that feature is used as a substitute; then, error classification verification is performed to compare the deviation of the current data from the historical average. According to the deviation range, three credibility levels are divided, and corresponding processing of direct use, weighted correction, and re-collection is performed respectively.

[0030] As a further technical solution of the present invention, the dominant disturbance source is identified based on a subset of the original features of the scene using a fuzzy C-means clustering algorithm: first, four types of features are fused, namely photovoltaic power output fluctuation rate, load factor, three-phase current deviation, and node sensitivity, to construct a sample feature vector; then, initial cluster centers for the three types of disturbances are preset, and the membership degree of the sample feature vector to each cluster center is calculated iteratively and the cluster centers are updated until the membership degree difference between two adjacent iterations meets the convergence accuracy requirement; finally, the disturbance type with the highest membership degree is determined as the dominant disturbance source of the current distribution network, which is divided into three categories: high fluctuation of new energy power output, peak load, and three-phase imbalance.

[0031] The multi-objective optimization weights include three categories of weights: voltage management, imbalance management, and resource utilization. The sum of the three categories of weights is a fixed value of 1. Based on the identified dominant disturbance type and node sensitivity, the proportion of each type of weight is adaptively and dynamically adjusted. The weights are updated synchronously at a fixed period that matches the frequency of changes in the distribution network conditions.

[0032] As a further technical solution of the present invention, the matching of scenario requirements and resource capabilities is calculated by integrating the matching scores of a single scenario and a single governance resource according to the preset weights of three dimensions: response matching degree, node matching degree, and cost matching degree. The resources are sorted from high to low and the top three resource combinations in the matching degree ranking are recommended. At the same time, an automatic resource failure replacement mechanism is set up. If the first-choice recommended resource suddenly goes offline or its adjustment capability fails, the matching score is automatically recalculated and the second-best recommended resource is switched.

[0033] As a further technical solution of the present invention, the multi-objective power configuration optimization uses dynamically allocated multi-objective optimization weights as weighting coefficients for each sub-optimization objective to construct a comprehensive optimization objective with voltage over-limit rate, three-phase imbalance, and resource utilization rate as the core; at the same time, it combines power injection constraints, resource adjustment margin constraints, and equipment safe operation constraints, and uses an algorithm combining genetic algorithm and particle swarm algorithm to iteratively solve the problem.

[0034] In the solution process, a genetic algorithm is first used to generate candidate power configuration solutions covering the entire constraint range. Then, a particle swarm optimization algorithm is used for iterative optimization. At the same time, a random disturbance correction strategy for the scenario of fluctuation in new energy output is added to avoid the algorithm getting trapped in local optima. The power configuration scheme is output after the deviation of the optimization objective between two adjacent iterations meets the convergence accuracy requirements. Finally, the scheme is substituted into three scenarios: sudden change in new energy output, large load fluctuation, and governance resource failure for disturbance resistance verification. If the governance requirements are not met, the optimization solution is retried.

[0035] As a further technical solution of the present invention, it also includes: building an electromagnetic transient simulation model that matches the actual operating parameters of the distribution network based on the output power configuration scheme, completing the verification of the governance effect under three operating conditions including large fluctuations in new energy output, peak load, and governance resource failure, and outputting a governance effect evaluation report; feeding back the verification results to iteratively correct the voltage-reactive power sensitivity matrix parameters, resource profile labels, feature extraction model weights, and optimization algorithm parameters.

[0036] Compared with existing technologies, the advantages of this multi-objective weighted voltage governance and power configuration method are:

[0037] By correcting the time dimension of equipment health assessment and the spatial dimension of topology coupling coefficient, the voltage-reactive power sensitivity matrix is ​​dynamically optimized in both time and space. This breaks through the limitations of traditional ideal equipment states and static models of fixed topologies, enabling the voltage-reactive power sensitivity matrix to reflect in real time the actual operating conditions of long-term performance degradation of the management equipment and dynamic changes in the distribution network topology. This fundamentally eliminates the deviation between theoretical adjustment parameters and actual field requirements, significantly improving the accuracy of voltage-reactive power correlation analysis. At the same time, based on the corrected voltage-reactive power sensitivity matrix, the sensitivity of nodes is quantified, and highly sensitive management nodes are accurately identified. This clarifies the target for priority allocation of management resources, ensuring the pertinence and effectiveness of voltage management from the perspective of basic data.

[0038] By precisely eliminating redundant features unrelated to voltage management, the purified feature dimensions are reduced compared to the original data, improving the efficiency of subsequent algorithm training, significantly reducing the computational load of the optimization algorithm, and enhancing the real-time response capability of the solution. At the same time, through a two-level trust verification mechanism based on blockchain traceability, a traceability tag that cannot be tampered with throughout its entire lifecycle is attached to the core feature data, and tamper verification and error classification verification are completed sequentially, effectively avoiding optimization decision deviations caused by tampered data and erroneous data, and ensuring the accuracy and reliability of subsequent optimization decisions.

[0039] By employing a fuzzy C-means clustering algorithm, this method accurately identifies three main sources of disturbance in the distribution network: high fluctuations in renewable energy output, peak loads, and three-phase imbalance. It then adaptively and dynamically allocates multi-objective optimization weights based on the dominant disturbance type and node sensitivity, overcoming the limitations of traditional fixed-weight optimization schemes that cannot adapt to different disturbance scenarios. This achieves real-time matching of optimization objective priorities with actual on-site governance needs. Differentiated weight adjustments for highly sensitive nodes and different disturbance scenarios ensure rapid voltage governance response in emergency situations while avoiding resource waste in routine scenarios. Furthermore, by quantifying the matching degree between scenario requirements and resource capabilities from three dimensions—response matching degree, node adaptability, and cost adaptability—it achieves precise matching of governance resources and governance scenarios, effectively avoiding low utilization and excessive governance costs caused by resource mismatch. Coupled with an automatic resource fault replacement mechanism, this ensures the continuity of governance scheduling, balancing governance effectiveness and operational economy.

[0040] When solving multi-objective power configuration problems, dynamically allocated optimization weights are used as the objective priority. Iterative optimization is completed by combining three types of hard constraints: power injection, resource margin, and equipment safety. At the same time, a random disturbance correction strategy is added for the high fluctuation scenario of new energy. When the algorithm gets stuck in a local optimum, the speed update formula is automatically adjusted, which effectively avoids the algorithm getting stuck in a local optimum and improves the global optimality and disturbance resistance of the power configuration scheme. Through disturbance resistance verification in disturbance scenarios, it is ensured that the scheme can still stably achieve the core objectives of voltage over-limit suppression and three-phase imbalance management under complex operating conditions such as sudden changes in new energy output, large load fluctuations, and resource management failures. This greatly improves the adaptability of the scheme to the complex operating environment of active distribution networks.

[0041] Through multi-scenario simulation verification and a full-process reverse feedback iteration mechanism, the verification results of the governance effect are pushed back to each preceding stage, realizing continuous iterative optimization of voltage-reactive power sensitivity matrix parameters, resource profile tags, feature extraction model weights, recommendation algorithms, and optimization parameters. This overcomes the limitation of traditional solutions being difficult to adapt to long-term changes in distribution network conditions after implementation. The solution can be directly adapted to the existing SOCADA system and governance equipment architecture of the distribution network without large-scale hardware modifications. At the same time, its adaptability can be continuously optimized as the distribution network operates, ensuring long-term stable power quality and operational economy of the distribution network. It has extremely strong engineering implementation value and long-term operational stability. Attached Figure Description

[0042] Figure 1 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] Please see Figure 1 The present invention provides an embodiment of a method for multi-objective weighted voltage governance and power configuration, comprising:

[0045] Acquire distribution network topology data, node electrical data, real-time distribution network topology information, and governance resource operation data; construct a voltage-reactive power sensitivity matrix based on distribution network topology data and node electrical data.

[0046] Based on the governance resource operation data and the real-time topology information of the distribution network, the voltage-reactive power sensitivity matrix is ​​corrected in both time and space dimensions. The node sensitivity is determined by the corrected voltage-reactive power sensitivity matrix, and the power injection constraints are determined based on the load type in the node electrical data. At the same time, a multi-dimensional resource profile is constructed based on the governance resource operation data.

[0047] Based on the corrected voltage-reactive power sensitivity matrix, nodal electrical data, and multi-dimensional resource profile, an original feature dataset is generated.

[0048] The original feature dataset is purified and its credibility is verified to obtain a feature dataset containing scene features and resource features.

[0049] Identify the dominant disturbance sources in the distribution network based on scenario characteristics, dynamically allocate multi-objective optimization weights based on the dominant disturbance sources and node sensitivity, and adapt and match scenario requirements with resource capabilities based on scenario characteristics and resource characteristics to generate resource combination recommendation results.

[0050] Using multi-objective optimization weights as the priority of optimization objectives and resource combination recommendation results as the configuration objects, combined with power injection constraints, multi-objective power configuration optimization is performed to generate power configuration schemes and output corresponding voltage management strategies.

[0051] Furthermore, in another embodiment provided, the distribution network topology data includes the electrical parameters of the distribution network lines, node connection relationships, and branch switch status;

[0052] The node electrical data includes the total number of distribution network nodes, node type, rated voltage, load type, and rated parameters;

[0053] Real-time distribution network topology information includes the real-time distribution network topology structure, line switching and switch operation information;

[0054] The governance resource operation data includes basic ledgers, real-time operation data, historical performance data, and equipment status information of governance equipment on the power grid side and user side.

[0055] The voltage-reactive power sensitivity matrix is ​​constructed using the Newton-Raphson power flow calculation method: First, the input distribution network topology data and node electrical data are standardized and preprocessed to form a power flow calculation dataset; initial values ​​for node voltage and phase angle are set for iteration, and the iteration convergence criterion is determined based on the node voltage deviation and phase angle deviation between adjacent iterations; the iteration loop is then entered: using the node voltage, phase angle, and the power flow calculation dataset as input, the power deviation between the actual power and the given power of each node is calculated; using the current node voltage and phase angle as input, the Jacobian matrix of the power flow equation is constructed, and the sub-blocks associated with reactive power-voltage are retained; the Jacobian matrix is ​​then used to calculate the power deviation between the actual power and the given power of each node. The Jacobian matrix is ​​inverted to obtain the correction amount for the node voltage and phase angle. The current node voltage and phase angle are updated with the correction amount, and the updated node voltage and phase angle are compared with the judgment criteria. If the node voltage deviation and phase angle deviation of two adjacent iterations do not meet the judgment criteria, the updated node voltage and phase angle are returned to the power deviation calculation step as the current value of the next iteration until the judgment criteria are met. Finally, the reactive power-voltage correlation sub-blocks retained in the converged Jacobian matrix are inversely operated to generate a voltage-reactive power sensitivity matrix. The matrix elements are used to characterize the voltage change amplitude of the corresponding node when a unit reactive power is injected into a single node.

[0056] Furthermore, in another embodiment, the time dimension correction is achieved by constructing an equipment health assessment model: based on the equipment status information in the governance resource operation data, three types of status indicators are extracted: equipment operating years, output current deviation rate, and real-time temperature rise, and an equipment health assessment model is constructed; corresponding weights are set according to the degree of influence of each indicator on equipment performance degradation, and the equipment health assessment model quantifies the degree of performance degradation of the equipment in long-term operation, and calculates the equipment health; then, the equipment health is used as an aging degradation factor to adapt and correct each element of the sensitivity matrix.

[0057] Spatial dimension correction is transformed into topology coupling correction, which is achieved through branch impedance coupling coefficient: First, the branch node group of the same line is identified through the real-time topology information of the distribution network, and the branch impedance coupling coefficient is calculated; then, the sensitivity matrix after time dimension correction is corrected a second time.

[0058] After completing the time and space dual-dimensional correction, based on the corrected voltage-reactive power sensitivity matrix, the absolute value of the voltage response intensity of each node to reactive power injection is calculated as the node sensitivity of that node. Nodes with node sensitivity exceeding the preset threshold are marked as high-sensitivity nodes and are used as the targets for priority allocation of subsequent governance resources. The preset threshold is set according to the voltage governance accuracy requirements of the distribution network.

[0059] Power injection constraints are set according to the characteristics of the load type. For resistive, inductive, and capacitive loads, the upper limit of active power injection, the range of reactive power adjustment, and the priority adjustment direction of the corresponding nodes are determined respectively. The power injection constraints and the modified sensitivity matrix together serve as the constraints for subsequent power configuration optimization.

[0060] Furthermore, in another embodiment, the multi-dimensional resource profile is constructed based on the basic ledger, real-time operation data, and historical performance data in the governance resource operation data, adopting a four-dimensional structured architecture of basic attributes, real-time operation status, historical performance, and adaptation scenarios. The basic attributes are generated based on the basic ledger and include fixed static data such as resource type, resource ID, access node, device rated capacity, and adjustable power factor range. The real-time operation status is obtained based on the real-time operation data and includes dynamically updated data on response speed, current adjustment margin, and operation status. The historical performance is generated based on historical performance data and includes full lifecycle statistics such as device adjustment accuracy, fault frequency, and fault-free operation time. The adaptation scenarios are preset tag data, including the disturbance type that the device prioritizes for adaptation, the sensitivity level of the adaptation node, and the cost threshold.

[0061] The current adjustment margin in the real-time operating status serves as the resource adjustment margin constraint for power configuration optimization; the rated capacity and adjustable power factor range of the equipment in the basic attributes serve as the equipment safe operation constraints for power configuration optimization.

[0062] The statistical data from historical performance and the tag data in the adapted scenarios constitute resource profile tags;

[0063] Establish a supporting dynamic update mechanism to update real-time operating status data at a preset fixed short cycle and iterate historical performance data at a preset fixed long cycle; the short cycle is set to 10 minutes based on the frequency of power distribution network condition monitoring, and the long cycle is set to monthly based on the accumulation cycle of equipment operation statistics.

[0064] Furthermore, in another embodiment provided, the original feature dataset is divided into two categories: a scenario original feature subset and a resource original feature subset. The scenario original feature subset is composed of the node sensitivity determined by the modified voltage-reactive power sensitivity matrix and the load type in the node electrical data. The resource original feature subset is composed of the full set of index data of the multi-dimensional resource profile.

[0065] Furthermore, in another embodiment, feature purification is achieved using an improved lightweight neural network: first, the original feature dataset is standardized and preprocessed; then, the improved lightweight neural network, after pre-training, extracts core scenario features and core resource features through parallel branches: the scenario feature branch extracts core scenario features related to the distribution network operating conditions, disturbance types, and governance requirements, and removes redundant scenario features that are not directly related to the voltage governance effect; the resource feature branch extracts core resource features related to the regulation capability, adaptability, and economy of governance equipment, and removes redundant resource features that are not directly related to resource scheduling; the improved lightweight neural network, after pre-training, serves as the feature extraction model.

[0066] Trustworthiness verification is achieved through blockchain distributed ledger technology: First, a traceability tag containing the collection device identifier, timestamp, digital signature, original data, and hash value is attached to each refined feature data; then, tamper verification is performed to compare the consistency between the current data and the blockchain evidence information. If they are inconsistent, historical compliant data of that feature is used as a substitute; finally, error classification verification is performed to compare the deviation of the current data from the historical average. According to the deviation magnitude, three trust levels are divided into three levels, and corresponding processing is performed for direct use, weighted correction, and re-collection, respectively.

[0067] Furthermore, in another embodiment, the dominant disturbance source is identified based on a subset of the original features of the scenario using a fuzzy C-means clustering algorithm: first, four types of features—photovoltaic power output volatility, load factor, three-phase current deviation, and node sensitivity—are fused to construct a sample feature vector; then, initial cluster centers for the three types of disturbances are preset, and the membership degree of the sample feature vector to each cluster center is calculated iteratively and the cluster centers are updated until the membership degree difference between two adjacent iterations meets the convergence accuracy requirement; finally, the disturbance type with the highest membership degree is determined as the dominant disturbance source of the current distribution network, and is divided into three categories: high fluctuation of new energy power output, peak load, and three-phase imbalance.

[0068] The multi-objective optimization weights include three categories: voltage management, imbalance management, and resource utilization. The sum of the three categories of weights is a fixed value of 1. Based on the identified dominant disturbance type and node sensitivity, the proportion of each category of weights is adaptively and dynamically adjusted. The weights are updated synchronously at a fixed period that matches the frequency of changes in the distribution network conditions.

[0069] Furthermore, in another embodiment, the matching of scenario requirements and resource capabilities is performed by calculating the matching score of a single scenario and a single governance resource based on preset weights of three dimensions: response matching degree, node matching degree, and cost matching degree. The resources are then sorted from high to low and the top three resource combinations with the highest matching degree are recommended. At the same time, an automatic resource failure replacement mechanism is set up. If the first-choice recommended resource suddenly goes offline or its adjustment capability fails, the matching score is automatically recalculated and the resource is switched to the second-best recommended resource.

[0070] Furthermore, in another embodiment provided, the multi-objective power configuration optimization uses dynamically allocated multi-objective optimization weights as weighting coefficients for each sub-optimization objective to construct a comprehensive optimization objective with voltage over-limit rate, three-phase imbalance, and resource utilization rate as the core; at the same time, it combines power injection constraints, resource adjustment margin constraints, and equipment safe operation constraints, and uses an algorithm combining genetic algorithm and particle swarm optimization to iteratively solve the problem.

[0071] In the solution process, a genetic algorithm is first used to generate candidate power configuration solutions covering the entire constraint range. Then, a particle swarm optimization algorithm is used for iterative optimization. At the same time, a random disturbance correction strategy for the scenario of fluctuation in new energy output is added to avoid the algorithm getting trapped in local optima. The power configuration scheme is output after the deviation of the optimization objective between two adjacent iterations meets the convergence accuracy requirements. Finally, the scheme is substituted into three scenarios: sudden change in new energy output, large load fluctuation, and governance resource failure for disturbance resistance verification. If the governance requirements are not met, the optimization solution is retried.

[0072] Furthermore, in another embodiment provided, the method further includes: building an electromagnetic transient simulation model that matches the actual operating parameters of the distribution network based on the output power configuration scheme, completing the verification of the governance effect under three operating conditions including large fluctuations in new energy output, peak load, and governance resource failure, and outputting a governance effect evaluation report; feeding back the verification results to iteratively correct the voltage-reactive power sensitivity matrix parameters, resource profile labels, feature extraction model weights, and optimization algorithm parameters.

[0073] An example application is provided:

[0074] Step 1: Basic Data Acquisition and Voltage-Reactive Power Sensitivity Matrix Construction:

[0075] 1. Collect four types of data:

[0076] Distribution network topology data: Electrical parameters of 10kV distribution network lines, including impedance parameters (resistance, reactance), inter-node connection relationships, and branch switch status;

[0077] Node electrical data: total number of distribution network nodes, node type (PQ node: load, user-side photovoltaic; PV node: grid-side photovoltaic; balance node: incoming line node), rated voltage of each node, load type and rated active or reactive power parameters;

[0078] Real-time distribution network topology information: Real-time distribution network topology, branch line switching status, and switch operation information acquired through the SOCADA data acquisition and monitoring control system;

[0079] Data on the operation of governance resources: Basic ledgers of governance equipment on the grid side (Static Var Generator (SVG), centralized energy storage) and the user side (residential PV, home energy storage) (data source for basic attributes of the profile, including resource type, resource ID, access node, rated capacity, adjustable power factor range), real-time operating parameters (including response speed, current adjustment margin, energy storage SOC, operating status), historical operating data (including adjustment accuracy, number of annual failures, mean time between failures, lifespan loss rate), and equipment status information (years of operation, output current deviation rate, real-time temperature rise).

[0080] 2. Construction of the base voltage-reactive power sensitivity matrix:

[0081] Using the Newton-Raphson power flow calculation method, a base voltage-reactive power sensitivity matrix reflecting the correlation between reactive power injection at a single node and voltage changes at all nodes in the network is constructed, specifically as follows:

[0082] Power flow calculation basic data preprocessing: Taking the collected distribution network topology data and node electrical data as input, the node connection relationship and line impedance are sorted out to form a standardized power flow calculation dataset;

[0083] Power flow iteration initialization: Set the initial phase angle of all nodes to 0 (the phase angle of the slack node remains constant), and the initial voltage value to the rated voltage (1.0 times the rated value); Determine the convergence criterion: the voltage deviation between adjacent iterations is less than 10. -5 Double the rated value, phase angle deviation less than 10 -5 When the radius is radian, the power flow is determined to be convergent;

[0084] Newton-Raphson iterative solution (until convergence):

[0085] Step 1: Calculate the actual power of each node using the current voltage and phase angle, and compare it with the given power (load power, photovoltaic output) to calculate the power deviation;

[0086] Step 2: Construct the Jacobian matrix of the power flow equation, focusing on retaining the sub-blocks related to reactive power and voltage, which will serve as the core for subsequent sensitivity calculations;

[0087] Step 3: Invert the Jacobian matrix and calculate the correction values ​​for the node voltage and phase angle;

[0088] Step 4: Update the node voltage and phase angle with the correction value, and repeat the above steps until the convergence criterion is met;

[0089] Base voltage-reactive power sensitivity matrix generation: Invert the reactive power-voltage sub-blocks in the converged Jacobian matrix to obtain the base voltage-reactive power sensitivity matrix. Each element in the matrix represents the voltage-reactive power correlation strength between the corresponding nodes, that is, the voltage change of the corresponding node when 1Mvar of reactive power is injected into a single node.

[0090] Step 2: Spatiotemporal dual maintenance of voltage-reactive power sensitivity matrix and construction of multi-dimensional resource profile:

[0091] 1. The spatiotemporal dual-maintenance voltage-reactive power sensitivity matrix addresses the shortcomings of traditional static models that do not consider equipment performance degradation and dynamic topology changes. It sequentially corrects the basic voltage-reactive power sensitivity matrix from both temporal and spatial dimensions to obtain a final voltage-reactive power sensitivity matrix that closely reflects the actual on-site operating conditions.

[0092] The time-dimensional (equipment aging) correction is based on three core status information: the operating years of the treatment equipment, the output current deviation rate, and the real-time temperature rise. An equipment health assessment model is constructed to quantify the performance degradation after long-term operation and to correct the base voltage-reactive power sensitivity matrix.

[0093]

[0094] In the formula, The health status of the k-th device is defined as [0.6, 1]. If the value is less than 0.6, the device is considered to need maintenance. The factor representing the impact of service life is... Let k be the number of years the equipment has been in operation. The output current deviation rate influence coefficient. Let K be the output current deviation rate of the k-th device; This represents the real-time temperature rise influence coefficient. This represents the real-time temperature rise of the kth device.

[0095] Using equipment health as the aging degradation factor, the base voltage-reactive power sensitivity matrix is ​​corrected element by element to obtain the time-dimension corrected voltage-reactive power sensitivity matrix. The formula is as follows:

[0096]

[0097]

[0098] In the formula, Let be the aging degradation factor of the k-th device. The base voltage-reactive power sensitivity matrix generated in step 1;

[0099] Spatial dimension (topology coupling) correction is based on real-time distribution network topology information, identifying node groups on the same 10kV branch, and calculating the branch impedance coupling coefficient. (The coupling coefficient for nodes on the same branch is 0.2-0.4, and for nodes on different branches it is 0). A second correction is performed on the time-dimension corrected voltage-reactive power sensitivity matrix to obtain the final voltage-reactive power sensitivity matrix after spatiotemporal dual correction. :

[0100]

[0101] The node sensitivity quantification is based on the final voltage-reactive power sensitivity matrix after spatiotemporal dual-maintenance correction, which quantifies the node sensitivity of each node: the node sensitivity is the response strength of the node's voltage to reactive power injection, that is, the absolute value of the corresponding element of the voltage-reactive power sensitivity matrix; all nodes are traversed, and nodes with voltage response strength exceeding the threshold are marked as high-sensitivity nodes, which are the targets for priority allocation of subsequent governance resources.

[0102] Branch node coupling coefficient The calculation is as follows:

[0103] For any two nodes on the same branch and Its branch node coupling coefficient is The calculation formula is:

[0104]

[0105]

[0106] In the formula, For nodes and The coupling coefficient of the branch node, For power to node Total equivalent reactance, For power to node Total equivalent reactance, From power source to node The total number of line segments, For the first branch The total reactance of a line segment is equal to the product of the reactance per unit length and the line length.

[0107] Power injection constraints are set separately based on the electrical characteristics of the three types of loads:

[0108] Resistive loads (residential lighting): Voltage is positively correlated with active power, and the active power injection is constrained to not exceed 15% of the rated active power of the load, and reactive power is preferentially injected as lagging reactive power;

[0109] Inductive load (industrial motor): The constraint is that the reactive power injection shall not exceed 1.2 times the rated reactive power of the load;

[0110] Capacitive load (compensation capacitor): The constraint is that the reactive power injection amount shall not exceed 50 kvar / node, and advanced reactive power shall be injected if necessary;

[0111] Power injection constraints and the modified sensitivity matrix are used together for subsequent power configuration optimization. The sensitivity matrix is ​​used to identify highly sensitive nodes and guide the priority allocation of governance resources, while the power injection constraints are used to limit the upper limit of active power injection and the range of reactive power adjustment for each node.

[0112] 2. Multi-dimensional Resource Profile Construction: Based on the governance resource operation data obtained in Step 1, a four-dimensional structured governance resource profile is constructed, consisting of basic attributes, real-time status, historical performance, and adaptation scenarios. The basic attributes, real-time operation status, and historical performance directly map to the basic ledger, real-time operation parameters, and historical operation data, respectively. The adaptation scenarios are pre-defined tag data, including the disturbance type that devices prioritize for adaptation, the sensitivity level of the adaptation node, and the cost threshold. These are configured by operations and maintenance personnel based on the resource type and historical governance effects. Specific dimensions and core indicators are as follows:

[0113]

[0114] Establish a dynamic data update mechanism: update real-time status data every 10 minutes and iterate historical performance data monthly to ensure that the profile data is consistent with the actual operating status of the device;

[0115] Step 3: Generation of the original feature dataset:

[0116] 1. Original feature subset of the scenario: covering all features related to the governance scenario, such as node sensitivity determined by the modified voltage-reactive power sensitivity matrix and load type in node electrical data;

[0117] 2. Subset of original resource features: Covers all features related to governance resource capabilities, such as response speed, adjustment accuracy, adjustment margin, unit adjustment cost, and applicable scenarios. All features are derived from the four-dimensional governance resource profile.

[0118] Step 4: Feature purification and credibility verification:

[0119] 1. Lightweight feature purification employs an improved lightweight convolutional neural network to perform feature distillation on the high-dimensional original feature dataset. Redundant features weakly correlated with voltage governance effects are removed, focusing on core governance features. This reduces the feature dimension compared to the original data, improving the efficiency of subsequent algorithm training. The specific process is as follows:

[0120] Data preprocessing: Standardize the original feature dataset to eliminate the differences in the units of different feature indicators and avoid interference from data of different magnitudes on the feature extraction results; at the same time, complete missing value completion and outlier removal to form a standardized feature input dataset;

[0121] Improved lightweight convolutional neural network architecture adaptation:

[0122] To address the high-dimensional and time-series characteristics of distribution network feature data, a lightweight convolutional neural network structure is adapted and improved:

[0123] The original network's depthwise separable convolutional core structure is retained, reducing the number of network parameters and computational load, thus adapting to the lightweight deployment requirements at the edge of the power distribution network.

[0124] A new feature attention branch is added, which assigns higher weights to feature channels related to voltage governance effects, weakens the interference of irrelevant features, and improves the extraction accuracy of core features.

[0125] Adjust the network output layer structure and set up two parallel output branches for scene features and resource features, respectively, to meet the purification requirements of the original scene feature set and the original resource feature set;

[0126] Branch feature extraction:

[0127] The preprocessed original feature dataset is input into the adapted and improved lightweight convolutional neural network, and feature extraction is performed through two parallel branches:

[0128] Scene feature branch: Through depthwise separable convolutional layers, extract core scene features related to distribution network conditions, disturbance types, and governance needs, and eliminate redundant scene features that are not directly related to voltage governance effects;

[0129] Resource Feature Branch: Through deep separable convolutional layers, core resource features related to the adjustment capability, adaptability, and economy of governance equipment are extracted, while redundant resource features that are not directly related to resource scheduling are eliminated;

[0130] Model training and feature output:

[0131] Model pre-training: Using 5000+ historical data points of "original features - governance effect", the improved lightweight convolutional neural network is trained. The training process uses Pearson correlation coefficient loss as the optimization objective to ensure that the extracted features are strongly correlated with the voltage governance effect. After training, the network weight parameters are fixed to form a standardized feature extraction model.

[0132] Core feature output: Input the preprocessed original feature dataset into the trained model, and output the purified scene core features and resource core features to form scene core feature library and resource core feature library respectively, thus completing feature purification;

[0133] 2. The full lifecycle trustworthiness verification employs blockchain traceability technology, assigning a traceability tag to each initial core feature containing five parts: data collection device ID, timestamp, digital signature, original data, and hash value. This ensures data is traceable and tamper-proof throughout its entire lifecycle, and performs two levels of verification sequentially:

[0134] Tamper verification: Automatically compare the current data with the hash value of the blockchain notarization. If they do not match, it is determined that the data has been tampered with, and the historical average data of that feature is immediately used to replace it.

[0135] Error grading verification: Calculate the deviation between the current data and the historical mean.

[0136]

[0137] In the formula, This is the current data to be verified. This data represents the historical average of this feature; it is then graded according to deviation values ​​to ensure the final data reliability is no less than 99%.

[0138] Data with a deviation of ≤5% is considered Grade A reliable data and can be used directly.

[0139] Data with a deviation of 5%-10% is classified as Grade B and used after weighted correction. The correction formula is: Corrected data = 0.7 +0.3· ;

[0140] Data with a deviation greater than 10% is classified as Grade C, triggering re-collection of edge nodes. If the deviation exceeds the standard multiple times, historical average data will be used instead.

[0141] After verification, a feature dataset containing trusted scene features and trusted resource features is generated.

[0142] Step 5: Identify the main disturbance source, assign dynamic weights, and match scenarios and resources:

[0143] 1. Dominant Disturbance Source Identification: Based on scene features in the feature dataset, fuzzy C-means clustering algorithm is used to identify the current dominant disturbance source in the distribution network. The specific process is as follows:

[0144] Sample feature fusion: Integrating four core features—PV output volatility, load factor, three-phase current deviation, and node sensitivity—to construct a 4-dimensional sample feature vector;

[0145] Cluster center initialization: Set the initial cluster centers for the three types of perturbations:

[0146] High PV volatility center: [0.25, 0.6, 0.03, 1.0] (high PV volatility, medium load factor, low three-phase deviation, high-sensitivity node);

[0147] Peak load center: [0.05, 0.9, 0.05, 0.7] (low PV volatility, high load factor, low-to-medium three-phase deviation, medium-sensitive node);

[0148] Three-phase imbalance center: [0.08, 0.7, 0.1, 0.8] (low-medium PV volatility, medium load factor, high three-phase deviation, medium-high sensitivity node);

[0149] Membership calculation and iterative convergence: Calculate the membership degree of the current sample to the three cluster centers (values ​​from 0 to 1, summing to 1), and repeatedly update the cluster centers and membership degrees until the difference in membership degrees between two adjacent iterations is less than 0.01 (convergence threshold), at which point iterative convergence is determined; the membership degree is calculated as follows:

[0150]

[0151] In the formula, For the first The sample belongs to the first Membership degree of a class The total number of cluster categories. For the first The feature vector of each sample For the first Cluster center vector of the class, For the first Cluster center vector of the class, For the first The feature vector of the i-th sample and the i-th sample The Euclidean distance between the cluster center vectors of the classes;

[0152] Dominant disturbance source determination: The disturbance type corresponding to the maximum membership degree of the sample is taken as the dominant disturbance source of the current distribution network, which is divided into three categories: "high fluctuation of photovoltaic power, peak load, and three-phase imbalance".

[0153] 2. Multi-objective optimization and dynamic weight allocation: Based on the identified dominant disturbance source and the node sensitivity determined by the modified voltage-reactive power sensitivity matrix, dynamic allocation of "voltage control" is performed. Imbalanced governance Resource utilization rate "The optimization weights for the three types of objectives are summed to 1, and the weights are updated every 5 minutes according to the perturbation."

[0154]

[0155] 3. Scene-Resource Adaptation Matching: Based on scene and resource features in the feature dataset, a matching score between scene and resource is quantitatively calculated using a weighted average of 40% for response matching, 30% for node adaptability, and 30% for cost adaptability. This generates a resource combination recommendation result. The matching score formula is as follows:

[0156]

[0157] In the formula, The matching score between the i-th scenario and the j-th resource is given. In response to the matching degree, For node adaptability, For cost-effectiveness;

[0158] The calculation rules for the fit of each dimension are as follows:

[0159] Response matching degree calculate

[0160] Step 1: Define the scene response requirement threshold (Emergency scenario) =0.5s, typical scenario =2s);

[0161] Step 2: Obtain resource response speed ;

[0162] Step 3: Calculate fitness: If : The faster the response time, the higher the score; if : If the response speed exceeds the threshold, the score decreases linearly, reaching a minimum of 0.1.

[0163] Node adaptability calculate

[0164] Step 1: Quantize node sensitivity (high sensitivity = 1.0, medium sensitivity = 0.7, low sensitivity = 0.4).

[0165] Step 2: Obtain resource adjustment precision ;

[0166] Step 3: Calculate fitness: Node sensitivity × ;

[0167] Cost fit calculate

[0168] Step 1: Define the upper limit of the scenario cost budget (Emergency scenario) =0.3 yuan / kvar·h, typical scenario =0.15 yuan / kvar·h);

[0169] Step 2: Obtain resource unit adjustment costs ;

[0170] Step 3: Calculate fitness: If : The lower the cost, the higher the score; if : If the cost exceeds the budget, the score will be fixed at the lowest value.

[0171] Sort by matching score from highest to lowest and output the top 3 resource combinations recommended; set up a fault backup mechanism: if the preferred resource suddenly fails, the model will automatically switch to the second-best resource within 0.1 seconds to ensure scheduling continuity;

[0172] Step 6: Multi-objective power configuration optimization and solution output:

[0173] 1. Construction of the multi-objective optimization function: Using the multi-objective optimization weights dynamically allocated in step 5 as the priority of the optimization objectives, the resource combination recommendation results as the configuration object, and combining power injection constraints, a multi-objective optimization function is constructed:

[0174]

[0175] In the formula, To comprehensively optimize the objectives, The weights for voltage management, imbalance management, and resource utilization are respectively assigned in step 5. Voltage over-limit rate, For three-phase imbalance, For resource utilization-related optimization terms (the higher the resource utilization, the better, so we take 1 - resource utilization to achieve minimization optimization).

[0176] Calculation formulas for each sub-objective:

[0177]

[0178]

[0179]

[0180] In the formula, the voltage over-limit time is the cumulative duration during which the node voltage exceeds the allowable deviation range within the statistical period, and the total time is the total duration of the statistical period. This represents the total number of distribution network nodes. For nodes The three-phase unbalance is calculated using the following formula:

[0181]

[0182]

[0183] In the formula, They are nodes The three-phase voltages of A, B, and C This is the average value of the three-phase voltage;

[0184] 2. Constraint Setting: The optimization process must satisfy three types of hard constraints:

[0185] Power injection constraint: Active power injection amount reactive power injection ,in This represents the maximum active power injection amount for the node. The reactive power adjustment range of the node;

[0186] Resource adjustment margin constraint: The actual adjustment amount of the recommended resource is less than or equal to its current adjustment margin;

[0187] Equipment safety operation constraints: energy storage charging and discharging power ≤ its maximum charging and discharging power, and photovoltaic inverter power factor within the range of [0.9, 1.0].

[0188] 3. The specific process of optimization and solution using the GA-PSO hybrid algorithm is as follows:

[0189] Initialization: Genetic algorithm (GA) is used to generate 100 power configuration candidate solutions, covering parameters such as SVG reactive power injection, energy storage charging and discharging power, and photovoltaic power factor, fully covering the entire constraint range;

[0190] Iterative search:

[0191] To address the high volatility of photovoltaic scenarios, a random perturbation factor is added to the GA crossover stage. The candidate solution is modified to improve its performance due to disturbances. The formula is as follows:

[0192]

[0193] in For the original candidate solutions, To correct the candidate solutions and enhance the robustness of the proposed solutions;

[0194] If the algorithm fails to significantly improve performance after three consecutive generations of optimization, it is determined to be trapped in a local optimum, and the velocity update formula for Particle Swarm Optimization (PSO) is adjusted:

[0195]

[0196] In the formula, For the updated particle velocity, The current particle velocity, For inertial weights, For acceleration coefficient, A random number in the interval [0,1]. This represents the average deviation of the global optimal solution from the previous five generations. This represents the optimal solution for an individual particle. This is the globally optimal solution for the population. This is the particle's current position vector (corresponding to the current power configuration scheme);

[0197] Convergence criterion: When the optimization target error is less than 3% and stable for 3 consecutive generations, the iteration converges and the initial power configuration scheme is output.

[0198] Unstable performance verification: Substitute the initial power configuration scheme into three disturbance scenarios: photovoltaic power drop of 20%, load increase of 15%, and resource failure replacement. If the voltage over-limit rate exceeds 0.5% or the imbalance exceeds 2.5%, the optimization is retried until the scheme meets the requirements.

[0199] 4. The final solution and strategy outputs a power configuration scheme that meets the multi-objective optimization objectives (example: SVG-01 injects 0.6Mvar reactive power, U-ES-15 energy storage device discharges 30kW, and photovoltaic power factor is adjusted to 0.92), and simultaneously outputs a multi-objective weighted voltage management strategy adapted to the current disturbance scenario;

[0200] Preferred options also include step 7: multi-scenario simulation verification and full-process closed-loop iteration.

[0201] 1. The distribution network simulation model is built based on actual distribution network parameters (line impedance, load type). Electromagnetic transient simulation software is used to build a simulation model that includes distributed photovoltaic power and recommended governance resources.

[0202] Photovoltaic grid connection settings: Prioritize connection to highly sensitive nodes, simulate the daily power curve (0-1000kW fluctuation from 6:00 AM to 6:00 PM), and superimpose ±15% random fluctuations to match the actual weather impact;

[0203] Resource allocation for governance: Access is based on the top 1 recommended resource combination, and the initial parameter values ​​are set to the power configuration scheme values ​​output in step 6;

[0204] Monitoring indicators: Real-time acquisition of four core indicators: voltage over-limit rate, three-phase imbalance, resource response delay, and governance cost;

[0205] 2. Multi-scenario governance effect verification: Simulation verification was completed for three types of scenarios, and a governance effect evaluation report was output, clarifying the correspondence between scenario type, recommended resources, power configuration, and governance effect.

[0206] Scenario 1: High fluctuation disturbance in photovoltaic power (at 12 noon, photovoltaic output suddenly drops from 1000kW to 600kW): The verification results are: SVG response delay 0.3s, energy storage response delay 1.1s, voltage over-limit rate reduced from 8% before optimization to 0.18%, three-phase imbalance reduced from 3.6% to 1.7%, and recommended resource utilization rate 92%;

[0207] Scenario 2: Peak load disturbance (7 PM, load rate reaches 90%): The verification result shows that the algorithm adjusts the reactive power injection of SVG by 0.4 Mvar, optimizes the three-phase output distribution of the photovoltaic inverter, reduces the three-phase imbalance from 4.2% to 1.9%, and the voltage over-limit rate is <0.1%.

[0208] Scenario 3: Recommended resource failure replacement (preferred SVG offline, automatic switching to the second-best SVG): The verification results showed a switching delay of 0.08s, the voltage over-limit rate briefly rose to 0.35% and then fell back to 0.22%, with no continuous over-limit;

[0209] 3. The full-process reverse feedback iteration pushes the simulation verification results back to all preceding steps, realizing full-process parameter iterative optimization and forming a complete closed loop of data generation - purification and verification - optimization solution - verification iteration:

[0210] Feedback to Step 1-2: Basic data iteration: Correct the voltage-reactive power sensitivity matrix coefficients according to the actual adjustment effect, update the governance resource profile labels according to the actual resource performance, and adjust the power injection constraint boundary according to the equipment operating status;

[0211] Feedback to Step 4: Feature Purification and Model Iteration: If the weight of a certain type of feature on the governance effect increases, adjust the feature extraction priority of the feature distillation model; if data from a certain type of equipment frequently exceeds the standard, tighten the error verification threshold.

[0212] Feedback to steps 5-6: Recommendation model and optimization algorithm iteration: Incorporate successful or failed cases into the collaborative filtering recommendation model training library and adjust the matching dimension weights; if the algorithm converges slowly, adjust the GA population size or PSO acceleration coefficient to achieve continuous optimization of the solution.

[0213] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for multi-objective weighted voltage management and power configuration, characterized in that, include: Acquire distribution network topology data, node electrical data, real-time distribution network topology information, and governance resource operation data; construct a voltage-reactive power sensitivity matrix based on distribution network topology data and node electrical data. Based on the governance resource operation data and the real-time topology information of the distribution network, the voltage-reactive power sensitivity matrix is ​​corrected in both time and space dimensions. The node sensitivity is determined by the corrected voltage-reactive power sensitivity matrix, and the power injection constraints are determined based on the load type in the node electrical data. Simultaneously construct multi-dimensional resource profiles based on governance resource operation data; Based on the corrected voltage-reactive power sensitivity matrix, nodal electrical data, and multi-dimensional resource profile, an original feature dataset is generated. The original feature dataset is purified and its credibility is verified to obtain a feature dataset containing scene features and resource features. Identify the dominant disturbance source in the distribution network based on scenario characteristics, and dynamically allocate multi-objective optimization weights based on the dominant disturbance source and node sensitivity; Based on scene features and resource features, the system adapts and matches scene requirements with resource capabilities to generate resource combination recommendation results. Using multi-objective optimization weights as the priority of optimization objectives and resource combination recommendation results as the configuration objects, combined with power injection constraints, multi-objective power configuration optimization is performed to generate power configuration schemes and output corresponding voltage management strategies.

2. The method for multi-objective weighted voltage management and power configuration according to claim 1, characterized in that, The distribution network topology data includes the electrical parameters of the distribution network lines, node connection relationships, and branch switch status; The node electrical data includes the total number of distribution network nodes, node type, rated voltage, load type, and rated parameters; The real-time topology information of the distribution network includes the real-time topology structure of the distribution network, line switching and switch operation information; The governance resource operation data includes basic ledgers, real-time operation data, historical performance data, and equipment status information of the governance equipment on the power grid side and the user side. The voltage-reactive power sensitivity matrix is ​​constructed using the Newton-Raphson power flow calculation method: first, the input distribution network topology data and node electrical data are standardized and preprocessed to form a power flow calculation dataset; then, the initial values ​​of node voltage and phase angle are set for iteration, and the iteration convergence criterion is determined based on the node voltage deviation and phase angle deviation between adjacent iterations. Enter the iterative loop: using the node voltage, phase angle and the power flow calculation dataset of the current iteration as input, calculate the power deviation between the actual power and the given power of each node; using the current node voltage and phase angle as input, construct the Jacobian matrix of the power flow equation, and retain the sub-blocks associated with reactive power and voltage; Inverting the Jacobian matrix yields the corrections for the node voltages and phase angles; The current node voltage and phase angle are updated with the correction amount, and the updated node voltage and phase angle are compared with the judgment criteria. If the node voltage deviation and phase angle deviation of two adjacent iterations do not meet the judgment criteria, the updated node voltage and phase angle are returned to the power deviation calculation step as the current value of the next iteration until the judgment criteria are met. Finally, the reactive power-voltage correlation sub-blocks retained in the converged Jacobi matrix are inversely operated to generate a voltage-reactive power sensitivity matrix. The matrix elements are used to characterize the voltage change amplitude of the corresponding node when a unit reactive power is injected into a single node.

3. The method for multi-objective weighted voltage management and power configuration according to claim 2, characterized in that, The time dimension correction is achieved by constructing an equipment health assessment model: based on the equipment status information in the governance resource operation data, three types of status indicators are extracted: equipment operating years, output current deviation rate, and real-time temperature rise, and an equipment health assessment model is constructed; corresponding weights are set according to the degree of influence of each indicator on equipment performance degradation, and the equipment health assessment model quantifies the degree of performance degradation of the equipment in long-term operation, and calculates the equipment health. Then, using the equipment health status as the aging and degradation factor, each element of the sensitivity matrix is ​​adapted and corrected. The spatial dimension correction is a topology coupling correction, which is achieved through the branch impedance coupling coefficient: first, the branch node group of the same line is identified through the real-time topology information of the distribution network, and the branch impedance coupling coefficient is calculated; then, the voltage-reactive power sensitivity matrix after the time dimension correction is corrected a second time. After completing the time and space dual-dimensional correction, based on the corrected voltage-reactive power sensitivity matrix, the absolute value of the voltage response intensity of each node to reactive power injection is calculated as the node sensitivity of that node. Nodes with node sensitivity exceeding a preset threshold are marked as high-sensitivity nodes and are prioritized for subsequent governance resource allocation. The power injection constraints are set according to the characteristics of the load type. For resistive, inductive, and capacitive loads, the upper limit of active power injection, the range of reactive power adjustment, and the priority adjustment direction of the corresponding nodes are determined respectively. The power injection constraints and the modified sensitivity matrix together serve as the constraints for subsequent power configuration optimization.

4. The method for multi-objective weighted voltage management and power configuration according to claim 2, characterized in that, The multi-dimensional resource profile is constructed based on the basic ledger, real-time operation data and historical performance data in the governance resource operation data, and adopts a four-dimensional structured architecture of basic attributes, real-time operation status, historical performance and adapted scenarios. The basic attributes are generated based on the basic ledger and include fixed static data such as resource type, resource ID, access node, device rated capacity, and adjustable power factor range. Real-time operating status is acquired based on the real-time operating data, including response speed, current adjustment margin, and dynamically updated operating status data; historical performance is generated based on historical performance data, including full lifecycle statistics of equipment adjustment accuracy, fault frequency, and fault-free operating time. The adaptation scenarios are based on preset tag data, including the disturbance types that the device prioritizes for adaptation, the sensitivity level of the adaptation nodes, and the cost threshold. The current adjustment margin in the real-time operating status serves as a resource adjustment margin constraint for power configuration optimization; the rated capacity and adjustable power factor range of the equipment in the basic attributes serve as equipment safe operation constraints for power configuration optimization. The statistical data in the historical performance and the tag data in the adapted scenario constitute the resource profile tag; Establish a supporting dynamic update mechanism to update real-time operating status data at a preset fixed short cycle and iterate historical performance data at a preset fixed long cycle; the short cycle is set to 10 minutes based on the frequency of power distribution network condition monitoring, and the long cycle is set to monthly based on the accumulation cycle of equipment operation statistics.

5. The method for multi-objective weighted voltage management and power configuration according to claim 2, characterized in that, The original feature dataset is divided into two categories: a subset of scene original features and a subset of resource original features. The original feature subset of the scenario consists of the node sensitivity determined by the modified voltage-reactive power sensitivity matrix and the load type in the node electrical data. The original feature subset of the resources consists of the integration of all index data of the multi-dimensional resource profile.

6. The method for multi-objective weighted voltage management and power configuration according to claim 1, characterized in that, The feature extraction is implemented using an improved lightweight neural network: first, the original feature dataset is standardized and preprocessed; then, the improved lightweight neural network, which has been pre-trained, extracts core scene features and core resource features through parallel branches. The scenario feature branch extracts core scenario features related to the power distribution network operating conditions, disturbance types, and governance needs, and removes redundant scenario features that are not directly related to the voltage governance effect. The resource feature branch extracts core resource features related to the adjustment capability, adaptability, and economy of the governance equipment, and eliminates redundant resource features that are not directly related to resource scheduling; the pre-trained improved lightweight neural network is used as the feature extraction model. The credibility verification is achieved through blockchain distributed ledger technology: First, a traceability tag containing the collection device identifier, timestamp, digital signature, original data, and hash value is attached to each purified feature data; then, tamper verification is performed to compare the consistency between the current data and the blockchain evidence information. If they are inconsistent, the historical compliant data of that feature is used as a substitute; then, error classification verification is performed to compare the deviation of the current data from the historical average. According to the deviation range, three credibility levels are divided, and corresponding processing of direct use, weighted correction, and re-collection is performed respectively.

7. The method for multi-objective weighted voltage management and power configuration according to claim 5, characterized in that, The dominant disturbance source is identified based on a subset of the original features of the scenario using a fuzzy C-means clustering algorithm: first, it integrates four types of features—PV power output volatility, load factor, three-phase current deviation, and node sensitivity—to construct a sample feature vector; then, it presets initial cluster centers for the three types of disturbances, iteratively calculates the membership degree of the sample feature vector to each cluster center, and updates the cluster centers until the membership degree difference between two adjacent iterations meets the convergence accuracy requirement; finally, the disturbance type with the highest membership degree is determined as the dominant disturbance source of the current distribution network, and is divided into three categories: high fluctuation of new energy power output, peak load, and three-phase imbalance. The multi-objective optimization weights include three categories of weights: voltage management, imbalance management, and resource utilization. The sum of the three categories of weights is a fixed value of 1. Based on the identified dominant disturbance type and node sensitivity, the proportion of each type of weight is adaptively and dynamically adjusted. The weights are updated synchronously at a fixed period that matches the frequency of changes in the distribution network conditions.

8. The method for multi-objective weighted voltage management and power configuration according to claim 1, characterized in that, The matching of scenario requirements and resource capabilities is performed by calculating the matching score of a single scenario and a single governance resource based on preset weights of three dimensions: response matching degree, node matching degree, and cost matching degree. The resources are then sorted from high to low and the top three resource combinations with the highest matching degree are recommended. At the same time, an automatic resource failure replacement mechanism is set up. If the first-choice recommended resource suddenly goes offline or its adjustment capability fails, the matching score is automatically recalculated and the second-best recommended resource is switched to.

9. A method for multi-objective weighted voltage management and power configuration according to claim 4, characterized in that, The multi-objective power configuration optimization uses dynamically allocated multi-objective optimization weights as weighting coefficients for each sub-optimization objective, constructing a comprehensive optimization objective with voltage over-limit rate, three-phase imbalance, and resource utilization rate as the core; at the same time, it combines power injection constraints, resource adjustment margin constraints, and equipment safe operation constraints, and uses an algorithm combining genetic algorithm and particle swarm optimization to solve iteratively. In the solution process, a genetic algorithm is first used to generate candidate power configuration solutions covering the entire constraint range. Then, a particle swarm optimization algorithm is used for iterative optimization. At the same time, a random disturbance correction strategy for the scenario of fluctuation in new energy output is added to avoid the algorithm getting trapped in local optima. The power configuration scheme is output after the deviation of the optimization objective between two adjacent iterations meets the convergence accuracy requirements. Finally, the scheme is substituted into three scenarios: sudden change in new energy output, large load fluctuation, and governance resource failure for disturbance resistance verification. If the governance requirements are not met, the optimization solution is retried.

10. A method for multi-objective weighted voltage management and power configuration according to claim 1, 4, or 6, characterized in that, Also includes: Based on the output power configuration scheme, an electromagnetic transient simulation model matching the actual operating parameters of the distribution network is built. The governance effect is verified under three operating conditions: large fluctuations in new energy output, peak load, and governance resource failure. A governance effect evaluation report is output. The verification results are fed back in reverse to iteratively correct the voltage-reactive power sensitivity matrix parameters, resource profile labels, feature extraction model weights, and optimize algorithm parameters.

Images